TLDR
- Overchat AI just added two new models: NVIDIA's Sana Video for text-to-video generation and Crystal Upscaler for AI-powered portrait enhancement.
- Sana Video generates a 5-second 720p clip in about 29 seconds using a Linear Diffusion Transformer architecture with linear attention — it was trained in just 12 days on 64 H100 GPUs, roughly 1% of the cost of comparable models.
- Sana supports resolutions up to 720×1280 and videos up to one minute long, with strong prompt adherence for simple scenes but limitations on complex physics or dynamic motion.
- Sana is best suited for b-roll, drone-style footage, and simple camera movements where speed and affordability matter more than flagship-level realism.
- Crystal Upscaler, powered by Clarity AI, performs 2× upscaling specifically optimized for portraits — turning a 720px image into 1440px while preserving likeness and enhancing skin texture.
- Crystal processes 1K images in around 1.2 seconds and 5K images in roughly 20 seconds, making AI avatars look more lifelike and restoring low-quality photos without losing natural features.
- Pricing on Overchat AI is 200 tokens per Sana Video generation and 20 tokens per Crystal Upscaler run.
- Both models are available now in Overchat AI's video and image generators, with Gemini 3 and Nano Banana 2 rumored to arrive soon.
We've summarized the main features of each model in a table below:
| Feature |
Sana Video |
Crystal Upscaler |
| Type |
Text-to-Video |
AI Image Upscaler |
| Best For |
Fast video generation of static scenes and stock footage |
Portrait enhancement, face details, skin texture |
| Resolution |
Up to 720×1280 |
2x upscale of the original |
| Generation Time |
~30 seconds for 5-second 720p video |
1–21 seconds depending on resolution |
| Cost |
20 credits |
200 credits |
| Key Strength |
Speed of generation |
Creating natural skin texture |
| Input |
Text prompts |
Low-resolution portraits and AI-generated faces |
What is Sana Video?
Sana Video is NVIDIA’s video generation model built on a Linear Diffusion Transformer architecture. It can generate videos at up to 720×1280 resolution and up to one minute in length — but its real strength lies in its speed, producing a full generation in under 30 seconds.
Fun fact: Sana Video was trained in just 12 days on 64 H100 GPUs, which is roughly 1% of the training cost of comparable models.
Key features:
- Generates a 5-second 720p video in about 29 seconds
- Achieves this speed through a linear attention mechanism
- Supports a text-to-video workflow
- Offers strong prompt adherence, though it can struggle with complex or dynamic scenes
In other words, this isn’t Sora 2 — you won’t get highly realistic physics or intricate motion. However, Sana Video is ideal for situations where speed and efficiency matter more than perfect realism: generating b-roll, drone-style footage, or simple camera movements. It’s both cheaper and faster than flagship models, making it a highly practical choice for quick video production.
What is Crystal Upscaler?
Crystal Upscaler is an AI image enhancer powered by Clarity AI, specifically optimized for upscaling portraits and faces — whether of real people or AI-generated characters.
The goal of the model is not only to enlarge and sharpen the image, but also to preserve likeness and enhance skin texture. That means plasticky, artificial-looking AI avatars become more lifelike, while low-quality photos of loved ones are carefully restored without losing their natural features.
The results are impressively photorealistic, with crisp details and balanced skin tones. In Overchat AI, the model performs 2× upscaling, so a 720px image on its longest side becomes 1440px after processing.
In terms of performance, Crystal Upscaler is fast — around 1.2 seconds for 1K images and roughly 20 seconds for 5K images.
Using it couldn’t be simpler: just upload the portrait you’d like to enhance — whether it’s an overly smooth AI render, a low-resolution selfie, or a blurry photo — then hit Generate. After a short processing time, you’ll get a sharp, natural-looking result. There’s no setup or manual editing required — it’s a straightforward upload and download workflow.
Cost and Pricing
Let's talk about how much these models cost to run on Overchat AI:
- Sana Video costs 200 tokens per generation
- Crystal Upscaler Cost costs 20 tokens per generation
Where to Access These Models
Both models are available in Overchat AI's video generator and image generator, respectively. Sign up for a free account to start using them.
- Start using Sana Video
- Start using Crystal Upscaler
Bottom Line
We just added two powerful new AI models to Overchat AI: Sana Video and Crystal Upscaler.
NVIDIA’s Sana is a text-to-video model known for its impressive speed, making it great for simple scenes with minimal motion. While it doesn’t quite match the quality of flagship models like Sora or Veo 3.1, its speed and affordability make it a highly practical option for quick video generation.
Crystal Upscaler is an AI-powered image enhancer that’s incredibly easy to use—just upload your image and click a single button. It’s specially optimized for portraits, perfect for sharpening blurry photos of people or improving the skin texture of AI-generated characters.
At Overchat AI, we’re constantly adding new models, so stay tuned for even more exciting updates in the near future—especially with Gemini 3 and Nano Banana 2 rumored to be just around the corner.
